Modeling Phase Transitions in Gene Expression State Space

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Successful quantitative approaches in biology have included building detailed local models or detecting robust signals in high-throughput data. In this proposal, both these methods are combined in an innovative way to study transcriptional changes in human tissue upon infection by oncogenic viruses. Such viruses can have a range of consequences, from minor changes to drastic transformations in the cell phenotype. Starting from a seed network consisting of all known information about the viral-host interaction, a Bayesian transcriptional network will be learned on the gene expression data. The Bayesian network is then transformed into an equivalent system of interacting electromagnetic spins. Examples of such spin systems have been studied in statistical physics, and they are known to have rich phase structures. The spin system corresponding to the host cell network will be simulated, and domains of aligned spins will be identified as genetic modules that characterize the response of the cell to perturbations. The activation levels of these modules will be used to demarcate phases in the gene expression state space. Novel phases and phase transitions discovered in this way will then be validated by experiments. This framework sifts out probabilistic interactions from noisy high- throughput data and then makes novel predictions based on the resulting network model. It is a new, quantitative, and biologically informative way to model perturbations to human cells. On a clinical level, it could be used to finely differentiate between various normal and disease states in patients, and to calculate which therapies would best reverse the progression of a disease. This technique has the potential to make medical diagnosis and treatment more efficient, directed and precise. PUBLIC HEALTH RELEVANCE: The goal of my research project is to quantify how perturbations to the human transcriptional network cause transitions between different phenotypes. Working in this framework, clinicians will be able to detect disease states using widely available high-throughput methods. They can then determine the personalized treatment, or combination of treatments, that will most efficiently reverse disease progression in a particular patient.